For decades, organisations have developed systems for managing people by defining roles, allocating responsibilities, establishing approval processes, reviewing performance and determining who is accountable when something goes wrong.
With AI agents increasingly more capable of completing multi-step tasks, organisations face a new management challenge. AI agents are not employees, of course, but if we increasingly delegate work to them, many familiar management questions suddenly become relevant. Who is now responsible for:
- Giving them instructions?
- Checking their work?
- Deciding how much autonomy should they have?
- Being accountable when something goes wrong?
Delegation is not abdication
AI may perform work, but it cannot assume organisational accountability.
If an AI-generated communication contains incorrect information, an automated process produces an inappropriate outcome or an agent overlooks an important requirement, responsibility does not disappear into the technology.
Organisations therefore need clear lines of accountability around who approves AI use, who monitors its performance, who checks its work and who responds when something goes wrong.
From using AI to managing AI
A traditional AI assistant responds to a request to draft an email, summarise a report or generate ideas, after which you review the response and decide what happens next.
AI agents potentially change this dynamic because, rather than completing a single task, they can be given a broader objective and complete a series of steps towards achieving it.
As that shift occurs, employees now need to define clear boundaries:
- What information can it access?
- What actions can it take?
- Which decisions can it make independently?
- When must it stop and ask for human input?
In other words, somebody needs to manage the delegation.
Who checks the work?
It would be insane to give an inexperienced employee an important task, provide no supervision and simply assume everything had been completed perfectly. Yet organisations risk doing exactly that by increasingly relying on AI.
The better AI becomes, the easier it may be to trust it, which makes oversight more complex rather than less important. If an AI agent completes 20 steps correctly but makes a serious error, will anyone notice?
Organisations need to move beyond simply saying there should be a “human in the loop” to determining who that human needs to be, and when and where they should specifically intervene:
- Should a person approve the plan before AI begins?
- Review particular decisions during the process?
- Check the final output?
- Only when the AI encounters something outside clearly defined boundaries?
A new workplace capability
In much the same way managers are trained to delegate their team to define the outcome, provide the necessary information, establish boundaries, clarify authority and review performance, the same must be done when working with AI. Employees need to decide:
- What should I do myself?
- What should I delegate to AI?
- How much autonomy should I allow?
- And where do I need to remain involved?
The future workplace involve managing employees using AI tools alongside managing work completed by AI or combinations of both. Which raises a question organisations should start considering now: If AI is going to do more of the work, who is responsible for managing how that work gets done?
At GAPS Professional Writing, our workplace AI training helps leaders and employees move beyond basic prompting to develop the judgement, delegation and oversight skills required to work effectively with increasingly capable AI.
Because the next AI skill may not simply be knowing how to use AI. It may be knowing how to manage it.